Using conversation MOPs in natural language interfaces

Abstract

Natural language interfaces must be easy to understand and flexible, if they are to be effective. To meet these goals many systems have focused on the conventional or intentional component of language. In this paper, we explore one method of combining convention and intention by representing conversation structure with conversation MOPs (Kellermann, Broeztmann, Lim, & Kitao, 1989; Schank, 1982). Conversation MOPs are schematic structures that store the rules of conversation as generalized episodes associated with the goals the episodes achieved. These associated goals can be used to allow the system to handle intention when necessary, while retaining the efficiency of being guided by convention. We discuss how conversation MOPs are used to process the opening portion of a dialogue in a natural language interface named JUDIS. JUDIS provides the interface for Julia, an interactive advice‐giving system.